Version 1.0

• Latent Dirichlet Allocation (LDA)  based topic model for scoring

• Single RESTful API

• Inputs –

• Output:

  • Overall score,

  • section scores of

    • skills

    • work experience,

    • education,

    • certification,

    • awards / interest,

    • resume drop,

    • steady job,

    • total work experience,

    • location score (score based on distance from job location)

    • resume raw scoring –  Entire JD vs Entire resume text

Version 2.0

• LDA Scoring Model – Normalization updates for final score

• Implemented Individual API split for all scoring component mentioned in the output of v1.0

• Enhancements for improving response time

• Moved Data processing to Parser

Version 2.1:

• Fixes and Handler for scoring a Bad job (Job description does not have more than 40 words)

• Fixes for LDA score  > 100

• Fixes for removing duplicate parsed data for each section of resume

Version 3.0:

• Implemented RoBERTa based Deep Learning Transformer model to give similarity score for resume components –  Skill score, Work experience score, Education score, Awards / Interests score, Certifications score, raw score

Version 3.1:

• Implemented LDA Score for Skill score, Certification score, education score

• Implemented RoBerta based transformer model similarity score for Work experience and Interests / Awards score

Version 4.0:

• Updated to RoBERTa model for all components (skill, work experience, education, awards/certifications, interests)

• Added weights to final scores of skill and work experience section

• Exploring further optimizations to improve performance speed

We are pleased to share the science and business logic of the Geescore™Jobseeker Scoring solution. It is a dynamic hybrid approach to scoring based on constant improvement; lowering bias, increasing objectivity, and score validation. Currently we use a hybrid:

• Deep Transformer model for work experience, skills, education, working domain and
• Classic scoring algorithms, with 3 levels of matching science, as well as real-world recruiting parameters, for derived attributes such as job matches, steady job, location, etc.